1. Field of Invention
The present invention relates to the field of video image detection, and particularly to a full-automatic detection method and system for static characteristic information in a dynamic image.
2. Description of Related Arts
In general, the static characteristic information in TV videos, such as station logos or other overlay icons and characters, always have different colors and frequency characteristics from video content. Therefore, the processing effect is probably far from satisfactory if an identical algorithm for the video content is adopted during post-processing. In addition, the high frequency characteristics of the icons always result in compression noise, such as mosquito-noise, during video compression or transmission process. On the other hand, the interference between static icon regions and video backgrounds may lead to erroneous estimation of image motion vectors around the icon regions during video de-interlacing operation.
In order to detect the overlay icons and characters, an existing detection method via calculating the inter-frame difference was available and seemed easy to operate due to the relatively-fixed positions of the overlay icons in the video. However, such detection method still has several disadvantages. For example, it seems impossible to obtain a comprehensive statistical results which contain the statistical data for all cases as sizes and shapes of the icons usually change significantly. In addition, some icons per se have different transparency degrees, which means that pixel values of the icons may vary with background pixel values, thus the inter-frame difference between the icons which have different background is no longer a fixed value. Therefore, it is difficult to detect opaque icons and icons with arbitrary transparency degrees by such detection method.
In the prior art, another icon detection method via edge detection is also used. This technology utilizes the characteristics that an icon region is stationary in time and space relative to a video region, takes video sequences as input, and detects changes in edges of an image to distinguish the icon region from a video content region. However, this method cannot automatically judge whether an icon region has been detected and thus cannot stop the detection operation automatically.